Data Riders: EDA Mini-Project

# Each team member would need to set their directory path to variable "project_dir"
# we will make sure everything else is relative to "project_dir"
project_dir <- "/home/admin-12/Documents/IMARTICUS/Data-Riders/EDA"
setwd(project_dir)

# Setup Packages
load_packages <- function () {
  # Imports
  packages <- c("VIM", "dplyr", "plotly", "psych", "corrplot", "cluster", "factoextra", "psych", "tm", "wordcloud2")
  installed_packages <- packages %in% rownames(installed.packages())
  if (any(installed_packages == FALSE)) {
    install.packages(packages[!installed_packages])
  }
  # Packages loading with suppressed messages
  suppressMessages(invisible(lapply(packages, library, character.only = TRUE)))
}
load_packages()

# Get the Raw Data from the file
read_file <- function () {
  data <- read.csv("data/cereals_data.csv")
  return (data)
}

# Handle missing values
handle_missing_values <- function (input) {
  output <- input
  output <- kNN(output)
  return (output)
}

# Clearn the data to make it good for processing
# eg. delete unwanted rows, change data types etc
data_processing <- function (input) {
  output <- input
  output <- handle_missing_values(output)

  # Set company names
  output$mfr_names <- as.character(output$mfr)
  output$mfr_names[output$mfr_names=="A"] <- "American Home Food Products"
  output$mfr_names[output$mfr_names=="G"] <- "General Mills"
  output$mfr_names[output$mfr_names=="K"] <- "Kelloggs"
  output$mfr_names[output$mfr_names=="N"] <- "Nabisco"
  output$mfr_names[output$mfr_names=="P"] <- "Post"
  output$mfr_names[output$mfr_names=="Q"] <- "Quaker Oats"
  output$mfr_names[output$mfr_names=="R"] <- "Ralston Purina"
  output$mfr_names_factor = as.factor(output$mfr_names)

  # Calorie Categories
  output <- within(output, {
    calories_category <- NA
    calories_category[calories>=50&calories<80] <- "L"
    calories_category[calories>=80&calories<110] <- "M"
    calories_category[calories>=110] <- "H"
  })

  # some data processing steps that would be required
  return (output)
}

# Get the Data Frame for the data
get_data <- function () {
  raw <- read_file()
  data <- data_processing(raw)
  return (data)
}

data <- get_data()
data_grp_mfr <- data %>% group_by(mfr_names_factor)

Part 1: Understanding Variables

Overall View

draw_density_plot <- function (value_set, xlab) {
  skewness <- round(skew(value_set), 2)
  mean <- round(mean(value_set), 2)
  median <- round(median(value_set), 2)
  title <- paste(xlab, "Density Chart   |  ", "Skewness:", skewness)
  plot(density(value_set), ylab = "Probabilty Density", xlab = xlab, main=title,)
  abline(v=mean, col="red")
  abline(v=median, col="blue")
}

Distribution of Calories (in the Products)

draw_density_plot(data$calories, "Calories")

Distribution of Ratings (for the Products)

draw_density_plot(data$rating, "Rating")

Part 2: Data Insights

Distribution of Rating vs Calorie Category

data %>% ggplot(aes(x = rating, fill = calories_category)) + geom_histogram() + scale_fill_brewer(palette = "Set1") +
        scale_x_continuous(name = "Ratings", expand = c(0,0)) +
        labs(fill = " Cal ", y ='Count', title = "Distribution of Rating vs Calorie Category", subtitle = "Higher the calories, Lower is the Rating") +
        theme_classic()

`stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Relationship of one variable with another

cor_result <- cor((data_grp_mfr %>% select_if(is.numeric))[-1])
corrplot(cor_result)

heatmap(cor_result)

good_bad_corr <- function (cor_result_filter) {
  #cor_result_filter <-
  cor_result_filter_names <- names(cor_result_filter)
  names(cor_result_filter) <- NULL
  cor_result_df <- data.frame(cor_result_filter_names, cor_result_filter) %>% arrange(cor_result_filter)
  names(cor_result_df) <- c("Parameters", "Corellation")
  ggplot(cor_result_df, aes(x=Parameters, y=Corellation)) + geom_bar(stat = "identity") +
          theme(axis.text.x=element_text(colour="black"), axis.text.y=element_text(colour="black"))
}

Influencers for Ratings

good_bad_corr(cor_result["rating",])

Influencers for Calories

good_bad_corr(cor_result["calories",])

Whats you get in one Serving

data %>% ggplot(aes(x = cups, fill = mfr_names)) +
        geom_histogram() +
        scale_x_continuous(name = "# of Cups", expand = c(0,0)) +
        labs(fill = "Manufacturer", title = "Distribution of Cups per Serving",
             subtitle = "different cups for servings", y="# of Products") +
        theme_classic() + scale_fill_brewer(palette="Dark2")

`stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

data %>% ggplot(aes(x = weight, fill = mfr_names)) + geom_histogram() +
    scale_x_continuous(name = "Weight (in ounces)", expand = c(0,0)) +
    labs(fill = "Manufacturer", title = "Distribution of Weight per Serving", subtitle = "Weight per serving") +
    theme_classic() + scale_fill_brewer(palette="Dark2")

`stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Best in the Market

d <- data %>% arrange(rating) %>% head(4) %>% select(name, rating)
ggplot(data=d, aes(x=name, y=rating)) + geom_bar(stat = "identity") +
        labs(y="Ratings") + scale_fill_brewer(palette="Dark2") +
        theme(axis.text.x=element_text(colour="black"), axis.text.y=element_text(colour="black"))

Part 3: Analysis by company

data_grp_mfr %>% plot_ly(y= ~rating, x= ~mfr_names_factor, type = "box", color= ~mfr_names_factor) %>%
  layout(title="Company wise Rating Consistency", xaxis=list(title="Manufactures"), yaxis=list(title="Rating"))

Part 4: Overall Picture

Word Cloud (error in markdown)

get_file_content <- function () {
  txt <- readLines("/home/admin-12/Documents/IMARTICUS/Data-Riders/EDA/data/health.txt")
  txt <- paste(txt, collapse = " ")
  txt <- tolower(txt)
  txt <- gsub("[^a-zA-Z]", " ", txt)
  txt <- gsub("\\s+", " ", txt)
  txt <- removeWords(txt, stopwords())
  words <- strsplit(txt, " ")
  word_freq <- table(words)
  return (word_freq)
}
word_freq <- get_file_content()
wordcloud2(word_freq, shape="star", size=10)

Feature Based Grouping (Clusters)

cluster_data <- data %>% select(calories, rating, protein, fat) %>% scale
row.names(cluster_data) <- data[,1]
km.res <- kmeans(cluster_data, 5, nstart = 25)
fviz_cluster(km.res, data=cluster_data, palette = "jco", ggtheme = theme_minimal())